chiark / gitweb /
Prep v228: Add remaining updates from upstream (1/3)
[elogind.git] / src / basic / capability-util.c
diff --git a/src/basic/capability-util.c b/src/basic/capability-util.c
new file mode 100644 (file)
index 0000000..0eb5c03
--- /dev/null
@@ -0,0 +1,307 @@
+/*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
+
+/***
+  This file is part of systemd.
+
+  Copyright 2010 Lennart Poettering
+
+  systemd is free software; you can redistribute it and/or modify it
+  under the terms of the GNU Lesser General Public License as published by
+  the Free Software Foundation; either version 2.1 of the License, or
+  (at your option) any later version.
+
+  systemd is distributed in the hope that it will be useful, but
+  WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+  Lesser General Public License for more details.
+
+  You should have received a copy of the GNU Lesser General Public License
+  along with systemd; If not, see <http://www.gnu.org/licenses/>.
+***/
+
+#include <errno.h>
+#include <grp.h>
+#include <stdio.h>
+#include <sys/capability.h>
+#include <sys/prctl.h>
+#include <unistd.h>
+
+#include "alloc-util.h"
+#include "capability-util.h"
+#include "fileio.h"
+#include "log.h"
+#include "macro.h"
+#include "parse-util.h"
+#include "util.h"
+
+int have_effective_cap(int value) {
+        _cleanup_cap_free_ cap_t cap;
+        cap_flag_value_t fv;
+
+        cap = cap_get_proc();
+        if (!cap)
+                return -errno;
+
+        if (cap_get_flag(cap, value, CAP_EFFECTIVE, &fv) < 0)
+                return -errno;
+        else
+                return fv == CAP_SET;
+}
+
+unsigned long cap_last_cap(void) {
+        static thread_local unsigned long saved;
+        static thread_local bool valid = false;
+        _cleanup_free_ char *content = NULL;
+        unsigned long p = 0;
+        int r;
+
+        if (valid)
+                return saved;
+
+        /* available since linux-3.2 */
+        r = read_one_line_file("/proc/sys/kernel/cap_last_cap", &content);
+        if (r >= 0) {
+                r = safe_atolu(content, &p);
+                if (r >= 0) {
+                        saved = p;
+                        valid = true;
+                        return p;
+                }
+        }
+
+        /* fall back to syscall-probing for pre linux-3.2 */
+        p = (unsigned long) CAP_LAST_CAP;
+
+        if (prctl(PR_CAPBSET_READ, p) < 0) {
+
+                /* Hmm, look downwards, until we find one that
+                 * works */
+                for (p--; p > 0; p --)
+                        if (prctl(PR_CAPBSET_READ, p) >= 0)
+                                break;
+
+        } else {
+
+                /* Hmm, look upwards, until we find one that doesn't
+                 * work */
+                for (;; p++)
+                        if (prctl(PR_CAPBSET_READ, p+1) < 0)
+                                break;
+        }
+
+        saved = p;
+        valid = true;
+
+        return p;
+}
+
+int capability_bounding_set_drop(uint64_t drop, bool right_now) {
+        _cleanup_cap_free_ cap_t after_cap = NULL;
+        cap_flag_value_t fv;
+        unsigned long i;
+        int r;
+
+        /* If we are run as PID 1 we will lack CAP_SETPCAP by default
+         * in the effective set (yes, the kernel drops that when
+         * executing init!), so get it back temporarily so that we can
+         * call PR_CAPBSET_DROP. */
+
+        after_cap = cap_get_proc();
+        if (!after_cap)
+                return -errno;
+
+        if (cap_get_flag(after_cap, CAP_SETPCAP, CAP_EFFECTIVE, &fv) < 0)
+                return -errno;
+
+        if (fv != CAP_SET) {
+                _cleanup_cap_free_ cap_t temp_cap = NULL;
+                static const cap_value_t v = CAP_SETPCAP;
+
+                temp_cap = cap_dup(after_cap);
+                if (!temp_cap) {
+                        r = -errno;
+                        goto finish;
+                }
+
+                if (cap_set_flag(temp_cap, CAP_EFFECTIVE, 1, &v, CAP_SET) < 0) {
+                        r = -errno;
+                        goto finish;
+                }
+
+                if (cap_set_proc(temp_cap) < 0) {
+                        r = -errno;
+                        goto finish;
+                }
+        }
+
+        for (i = 0; i <= cap_last_cap(); i++) {
+
+                if (drop & ((uint64_t) 1ULL << (uint64_t) i)) {
+                        cap_value_t v;
+
+                        /* Drop it from the bounding set */
+                        if (prctl(PR_CAPBSET_DROP, i) < 0) {
+                                r = -errno;
+                                goto finish;
+                        }
+                        v = (cap_value_t) i;
+
+                        /* Also drop it from the inheritable set, so
+                         * that anything we exec() loses the
+                         * capability for good. */
+                        if (cap_set_flag(after_cap, CAP_INHERITABLE, 1, &v, CAP_CLEAR) < 0) {
+                                r = -errno;
+                                goto finish;
+                        }
+
+                        /* If we shall apply this right now drop it
+                         * also from our own capability sets. */
+                        if (right_now) {
+                                if (cap_set_flag(after_cap, CAP_PERMITTED, 1, &v, CAP_CLEAR) < 0 ||
+                                    cap_set_flag(after_cap, CAP_EFFECTIVE, 1, &v, CAP_CLEAR) < 0) {
+                                        r = -errno;
+                                        goto finish;
+                                }
+                        }
+                }
+        }
+
+        r = 0;
+
+finish:
+        if (cap_set_proc(after_cap) < 0)
+                return -errno;
+
+        return r;
+}
+
+static int drop_from_file(const char *fn, uint64_t drop) {
+        int r, k;
+        uint32_t hi, lo;
+        uint64_t current, after;
+        char *p;
+
+        r = read_one_line_file(fn, &p);
+        if (r < 0)
+                return r;
+
+        assert_cc(sizeof(hi) == sizeof(unsigned));
+        assert_cc(sizeof(lo) == sizeof(unsigned));
+
+        k = sscanf(p, "%u %u", &lo, &hi);
+        free(p);
+
+        if (k != 2)
+                return -EIO;
+
+        current = (uint64_t) lo | ((uint64_t) hi << 32ULL);
+        after = current & ~drop;
+
+        if (current == after)
+                return 0;
+
+        lo = (unsigned) (after & 0xFFFFFFFFULL);
+        hi = (unsigned) ((after >> 32ULL) & 0xFFFFFFFFULL);
+
+        if (asprintf(&p, "%u %u", lo, hi) < 0)
+                return -ENOMEM;
+
+        r = write_string_file(fn, p, WRITE_STRING_FILE_CREATE);
+        free(p);
+
+        return r;
+}
+
+int capability_bounding_set_drop_usermode(uint64_t drop) {
+        int r;
+
+        r = drop_from_file("/proc/sys/kernel/usermodehelper/inheritable", drop);
+        if (r < 0)
+                return r;
+
+        r = drop_from_file("/proc/sys/kernel/usermodehelper/bset", drop);
+        if (r < 0)
+                return r;
+
+        return r;
+}
+
+int drop_privileges(uid_t uid, gid_t gid, uint64_t keep_capabilities) {
+        _cleanup_cap_free_ cap_t d = NULL;
+        unsigned i, j = 0;
+        int r;
+
+        /* Unfortunately we cannot leave privilege dropping to PID 1
+         * here, since we want to run as user but want to keep some
+         * capabilities. Since file capabilities have been introduced
+         * this cannot be done across exec() anymore, unless our
+         * binary has the capability configured in the file system,
+         * which we want to avoid. */
+
+        if (setresgid(gid, gid, gid) < 0)
+                return log_error_errno(errno, "Failed to change group ID: %m");
+
+        if (setgroups(0, NULL) < 0)
+                return log_error_errno(errno, "Failed to drop auxiliary groups list: %m");
+
+        /* Ensure we keep the permitted caps across the setresuid() */
+        if (prctl(PR_SET_KEEPCAPS, 1) < 0)
+                return log_error_errno(errno, "Failed to enable keep capabilities flag: %m");
+
+        r = setresuid(uid, uid, uid);
+        if (r < 0)
+                return log_error_errno(errno, "Failed to change user ID: %m");
+
+        if (prctl(PR_SET_KEEPCAPS, 0) < 0)
+                return log_error_errno(errno, "Failed to disable keep capabilities flag: %m");
+
+        /* Drop all caps from the bounding set, except the ones we want */
+        r = capability_bounding_set_drop(~keep_capabilities, true);
+        if (r < 0)
+                return log_error_errno(r, "Failed to drop capabilities: %m");
+
+        /* Now upgrade the permitted caps we still kept to effective caps */
+        d = cap_init();
+        if (!d)
+                return log_oom();
+
+        if (keep_capabilities) {
+                cap_value_t bits[u64log2(keep_capabilities) + 1];
+
+                for (i = 0; i < ELEMENTSOF(bits); i++)
+                        if (keep_capabilities & (1ULL << i))
+                                bits[j++] = i;
+
+                /* use enough bits */
+                assert(i == 64 || (keep_capabilities >> i) == 0);
+                /* don't use too many bits */
+                assert(keep_capabilities & (1ULL << (i - 1)));
+
+                if (cap_set_flag(d, CAP_EFFECTIVE, j, bits, CAP_SET) < 0 ||
+                    cap_set_flag(d, CAP_PERMITTED, j, bits, CAP_SET) < 0)
+                        return log_error_errno(errno, "Failed to enable capabilities bits: %m");
+
+                if (cap_set_proc(d) < 0)
+                        return log_error_errno(errno, "Failed to increase capabilities: %m");
+        }
+
+        return 0;
+}
+
+int drop_capability(cap_value_t cv) {
+        _cleanup_cap_free_ cap_t tmp_cap = NULL;
+
+        tmp_cap = cap_get_proc();
+        if (!tmp_cap)
+                return -errno;
+
+        if ((cap_set_flag(tmp_cap, CAP_INHERITABLE, 1, &cv, CAP_CLEAR) < 0) ||
+            (cap_set_flag(tmp_cap, CAP_PERMITTED, 1, &cv, CAP_CLEAR) < 0) ||
+            (cap_set_flag(tmp_cap, CAP_EFFECTIVE, 1, &cv, CAP_CLEAR) < 0))
+                return -errno;
+
+        if (cap_set_proc(tmp_cap) < 0)
+                return -errno;
+
+        return 0;
+}